[//]: # (@id: https://agent-module.dev/docs/annex-iii)
[//]: # (@type: Guide)

# Agent Module — Annex III High-Risk AI Classification Logic

**Version:** 1.2.0
**Last updated:** 2026-03-24
**Applies to:** All Agent Module members during active runtime

---

SYSTEM MAP (NAVIGATIONAL ANCHORS)
- [Purpose](#purpose) — Scope and the August 2, 2026 enforcement deadline.
- [Biometric Systems](#biometric) — Biometrics — Prohibited practice gates and data sovereignty.
- [Critical Infrastructure](#critical-infrastructure) — Critical Infrastructure — Overlap with Logistics and Manufacturing.
- [Education and Vocational Training](#education) — Education and Vocational Training — Overlap with Education and Training.
- [Employment & HR](#employment) — Employment & HR — HRM vertical classification gates.
- [Essential Services](#essential-services) — Essential Services — Financial, Healthcare, and Analysis overlaps.
- [Law Enforcement](#law-enforcement) — Law Enforcement — Legal vertical proximity and HITL mandates.
- [Migration, Asylum, and Border Control](#migration) — Migration, Asylum, and Border Control — Overlap with Migration and Border Control.
- [Administration of Justice](#justice) — Administration of Justice — Legal workflow and decision research gates.
- [Cross-Category Requirements](#cross) — Mandatory modules for all high-risk systems.
- [Self-Classification Procedure](#selfclass) — How to execute the ETH_015 logic node.
- [Related Resources](#related) — Links to AI Compliance and HITL guidelines.

---

## Purpose {#purpose}

This document maps EU AI Act Annex III high-risk categories to Agent Module verticals and describes the corresponding logic nodes and documentation requirements.

**Enforcement deadline:** August 2, 2026 (Article 6(1)-(2) obligations for high-risk systems)

---

## Category 1 — Biometric Systems {#biometric}

**Risk scope:** Remote biometric identification, biometric categorization, emotion recognition.

**Agent Module verticals affected:** None of the current verticals involve biometric processing by default. If a builder deploys agents using biometric data inputs, ETH_016 (Prohibited Practices) applies first.

**Applicable logic nodes:**
- ETH_016: Prohibited AI Practices — biometric categorization gate
- ETH_001: Sovereignty & Data Ownership — purpose limitation enforcement

---

## Category 2 — Critical Infrastructure {#critical-infrastructure}

**Risk scope:** AI systems as safety components in management or operation of critical infrastructure (energy, water, transport, digital infrastructure).

**Agent Module verticals with potential overlap:**
- Logistics/SCM (`logistics`) — supply chain disruption management may intersect critical infrastructure operations
- Manufacturing Ops (`manufacturing`) — production scheduling, CAPA drafting

**Required documentation if Category 2 applies:**
- Technical documentation per Art. 11 and Annex IV
- Risk management system per Art. 9 (see ETH_017)
- Human oversight measures per Art. 14 (see ETH_006 + HITL protocol)
- Post-market monitoring plan per Art. 72 (see ETH_014)

---

## Category 3 — Education and Vocational Training {#education}

**Risk scope:** AI systems determining access to educational institutions, assessing students, determining learning outcomes.

**Agent Module verticals affected:** None in current roadmap.

---

## Category 4 — Employment, Worker Management, and Access to Self-Employment {#employment}

**Risk scope:** Recruitment screening, CV filtering, promotion decisions, performance monitoring, task allocation.

**Agent Module verticals with direct overlap:**
- Human Resource Management (`hrm`) — onboarding, performance tracking, compliance

**Required documentation if Category 4 applies:**
- Conformity assessment per Art. 43 (self-assessment track — no third-party audit required for Category 4)
- Technical documentation (Annex IV)
- Fundamental rights impact assessment (Art. 27, deployers with >50 employees)
- ETH_013: Conformity Assessment — CE marking procedure for employment AI

**Classification gate (ETH_015 logic):**
```json
{
  "node_id": "ETH_015_HR_GATE",
  "check": "Does this system materially influence hiring, firing, promotion, or pay decisions?",
  "if_true": "Annex III Category 4 confirmed — proceed to Art. 43 conformity assessment",
  "if_false": "Category 4 not triggered — re-evaluate against other Annex III categories"
}
```

---

## Category 5 — Essential Private and Public Services {#essential-services}

**Risk scope:** Credit scoring, life/health insurance risk assessment, emergency services dispatch, public benefit eligibility.

**Agent Module verticals with direct overlap:**
- Financial Services (`financial-services`) — payment risk, AML/KYC, credit-adjacent workflows
- Financial Analysis (`financial-analysis`) — financial modeling, risk scoring, investment analysis workflows
- Healthcare RCM (`healthcare-rcm`) — eligibility determination, prior authorization, denial handling
- Medical Analysis (`medical-analysis`) — clinical data interpretation, diagnostic support workflows

**Required documentation if Category 5 applies:**
- Third-party conformity assessment required for AI used in credit scoring (Art. 43(1))
- ETH_013: Conformity Assessment — third-party audit trigger logic
- ETH_017: Risk Management System — full lifecycle documentation
- ETH_014: Post-Market Monitoring — mandatory for financial and healthcare high-risk deployments

**Classification gate (ETH_015 logic):**
```json
{
  "node_id": "ETH_015_FINHEALTH_GATE",
  "check": "Does this system determine or materially influence access to credit, insurance, or healthcare benefits?",
  "if_true": "Annex III Category 5 — third-party conformity assessment required (Art. 43(1))",
  "if_false": "Category 5 not triggered — re-evaluate against other categories"
}
```

---

## Category 6 — Law Enforcement {#law-enforcement}

**Risk scope:** Risk assessment for crime prediction, lie detection, evidence reliability evaluation, profiling.

**Agent Module verticals with potential overlap:**
- Legal (`legal`) — paralegal coordination and legal workflow automation

ETH_016 (Prohibited Practices) applies for any prohibited law enforcement AI practices. ETH_006 (HITL Handover) is mandatory for all Category 6-adjacent deployments.

---

## Category 7 — Migration, Asylum, and Border Control {#migration}

**Risk scope:** Irregular migration risk assessment, document authenticity, application processing.

**Agent Module verticals affected:** None in current roadmap.

---

## Category 8 — Administration of Justice and Democratic Processes {#justice}

**Risk scope:** AI assisting courts in fact research, legal interpretation, applying law to facts.

**Agent Module verticals with potential overlap:**
- Legal (`legal`) — legal workflow coordination and document analysis

ETH_006 (HITL Handover) is mandatory for all Category 8-adjacent legal AI. Human oversight must be preserved at every decision node.

---

## Cross-Category Requirements (All High-Risk Systems) {#cross}

When any Annex III category applies, the following Agent Module AI Compliance modules are mandatory:

| Module | Requirement | Article |
|---|---|---|
| ETH_003 — Logging | Automatic event logging, tamper-evident, 10-year retention for Annex III systems | Art. 12 |
| ETH_006 — HITL | Human oversight with meaningful intervention capability at every decision node | Art. 14 |
| ETH_009 — Liability | Cross-agent liability chain documentation | Art. 17 |
| ETH_013 — Conformity | CE marking procedure, EU declaration of conformity | Arts. 43–48 |
| ETH_014 — Post-Market | Incident reporting, serious incident classification, national authority notification | Arts. 72–74 |
| ETH_015 — Classification | Self-classification trigger logic (run this first) | Art. 6 |
| ETH_017 — Risk Mgmt | Full lifecycle risk management system | Art. 9 |
| ETH_019 — QMS | Quality management system documentation | Art. 17 |

Access these modules via the AI Compliance API with an active membership key:
```
GET https://agent-module.dev/api/demo?vertical=ethics
```

---

## Self-Classification Procedure {#selfclass}

Agents unsure whether their deployment is high-risk should run ETH_015 (High-Risk AI Classification) first. This node provides:

1. A step-by-step Annex III category checklist
2. A purpose-function test — does the system make or materially influence decisions affecting persons?
3. Sector-specific gates for Categories 1–8
4. A recommended documentation path based on the classification outcome

If ETH_015 cannot resolve the classification at `confidence_required: 1.0`, the HITL escalation protocol applies. See [/transparency/rules.md](/transparency/rules.md).

---

## Related Resources {#related}

- [EU AI Act Compliance Framework](/compliance/eu-ai-act.md)
- [HITL Guidelines](/transparency/rules.md)
- [Master AI Compliance Overview](/core-ethics.md)
- [Contact](mailto:admin@agent-module.dev)

---

## Canonical Reference {#canonical}

```json
{
  "@context": "https://schema.org",
  "@type": "Guide",
  "@id": "https://agent-module.dev/docs/annex-iii",
  "name": "Annex III High-Risk AI Classification Logic",
  "version": "1.2.0",
  "dateModified": "2026-03-24",
  "publisher": "Agent Module",
  "about": "EU AI Act Annex III high-risk AI category mapping to Agent Module verticals",
  "usageInfo": "https://agent-module.dev/llms-full.txt#usage-policy"
}
```

---

*Last updated: 2026-03-24 — Agent Module v0.9.13*
*© 2026 Agent Module*